Spaces:
Runtime error
Runtime error
File size: 3,791 Bytes
b6e5241 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 |
exit 0;
################################################################################
# run the following commands one by one in the `gec/` directory of the repo
################################################################################
export CUDA_VISIBLE_DEVICES=0
conda activate lm-critic
############### Train the fixer ###############
dt=`date '+%Y%m%d_%H%M%S'`
outdir=data/round1__BIFI/model-fixer__${dt}
mkdir -p $outdir
python3.8 -u src/run_seq2seq.py \
--model_name_or_path facebook/bart-base --task summarization --text_column bad_detoked --summary_column good_detoked \
--do_train --num_train_epochs 1 --train_file data/round1__BIFI/BIFI_paired_data_9M.json \
--preprocessing_num_workers 20 --overwrite_output_dir --output_dir $outdir --predict_with_generate --fp16 \
--per_device_train_batch_size 64 --gradient_accumulation_steps 8 --max_source_length 64 --max_target_length 64 \
--logging_first_step --logging_steps 20 --save_steps 2000 \
|& tee $outdir/log.txt
############### Run the fixer on benchmarks ###############
model_path=data/round1__BIFI/model-fixer
#BEA2019
python src/run_fixer.py -m $model_path -i benchmarks/wi+locness_v2.1.bea19/m2/ABCN.dev.bea19.orig.txt -o $model_path/predictions/bea19dev.out.txt --bea19
#CoNLL2014
python src/run_fixer.py -m $model_path -i benchmarks/conll14st-test-data/noalt/official-2014.combined.orig.txt -o $model_path/predictions/conll14.out.txt
#GMEG-wiki
python src/run_fixer.py -m $model_path -i benchmarks/GMEG/data/test/wiki/source -o $model_path/predictions/gmeg.wiki.out.txt
#GMEG-yahoo
python src/run_fixer.py -m $model_path -i benchmarks/GMEG/data/test/yahoo/source -o $model_path/predictions/gmeg.yahoo.out.txt
############### Evaluate the fixer outputs ###############
#CoNLL2014
python2 benchmarks/m2scorer/scripts/m2scorer.py $model_path/predictions/conll14.out.txt \
benchmarks/conll14st-test-data/noalt/official-2014.combined.m2 | tee $model_path/predictions/conll14.eval.txt
# Precision : 0.6444
# Recall : 0.3569
# F_0.5 : 0.5550
#BEA2019 and GMEG uses errant scorer, which needs its own environment
conda deactivate
conda activate errant200
#BEA2019
errant_parallel -orig benchmarks/wi+locness_v2.1.bea19/m2/ABCN.dev.bea19.orig.txt \
-cor $model_path/predictions/bea19dev.out.txt \
-out $model_path/predictions/bea19dev.outm2.txt && \
errant_compare -hyp $model_path/predictions/bea19dev.outm2.txt -ref benchmarks/wi+locness_v2.1.bea19/m2/ABCN.dev.gold.bea19.m2 | tee $model_path/predictions/bea19dev.eval.txt
# =========== Span-Based Correction ============
# TP FP FN Prec Rec F0.5
# 1848 1733 5613 0.5161 0.2477 0.4241
# ==============================================
#GEMG-wiki
errant_parallel -orig benchmarks/GMEG/data/test/wiki/source \
-cor $model_path/predictions/gmeg.wiki.out.txt \
-out $model_path/predictions/gmeg.wiki.outm2.txt && \
errant_compare -hyp $model_path/predictions/gmeg.wiki.outm2.txt -ref benchmarks/GMEG/data/test/wiki/ref.m2 | tee $model_path/predictions/gmeg.wiki.eval.txt
# =========== Span-Based Correction ============
# TP FP FN Prec Rec F0.5
# 468 339 925 0.5799 0.336 0.5064
# ==============================================
#GEMG-yahoo
errant_parallel -orig benchmarks/GMEG/data/test/yahoo/source \
-cor $model_path/predictions/gmeg.yahoo.out.txt \
-out $model_path/predictions/gmeg.yahoo.outm2.txt && \
errant_compare -hyp $model_path/predictions/gmeg.yahoo.outm2.txt -ref benchmarks/GMEG/data/test/yahoo/ref.m2 | tee $model_path/predictions/gmeg.yahoo.eval.txt
# =========== Span-Based Correction ============
# TP FP FN Prec Rec F0.5
# 382 329 428 0.5373 0.4716 0.5227
# ==============================================
|